Showing posts with label AI-generated images. Show all posts
Showing posts with label AI-generated images. Show all posts

Wednesday, October 7, 2026

When AI Draws Me a Dog I Never Had (And What It Means)

I tried a trending prompt: 

  "Create a caricature of me based on what you know from our chats."

Most of it was spot on: my work, my interests, my creator life. Then I noticed a dog in the corner, labeled as my stress buster.



I don't own a dog, and I've never talked to the AI about one. When I asked where it came from, the answer was basically: it seemed like a fitting detail.





What happened

The AI didn't look up a hidden fact about me. It filled a gap with a plausible invention.

Caricatures have a pattern. They usually include a work setup, a signature trait, and a hobby or stress relief. The AI had my work and interests but nothing for the last slot, so it made one up.

It predicts what's likely, not what's true. Image models and chatbots generate the most statistically plausible completion. For "stress buster", a dog is one of the most common answers in the data they learned from.

Confidence isn't accuracy. The dog was drawn with the same polish as the correct details, so nothing signaled that it was a guess.

Chat history is a thin sketch. AI only has fragments of what you've said. Where the fragments run out, it improvises.

This is the same behavior behind text "hallucinations", just in visual form.

Why it matters

Fun trends like this feel personal, so it's easy to treat the output as insight. But a made-up detail looks the same as a real one. If I'd posted it without checking, people would have assumed I own a dog.

The same thing happens in more serious work: invented statistics, fake citations, or tool features that don't exist.

How to get more accurate results

Set boundaries.  Add: "Only use details I've actually told you. Leave anything else out."

Ask for a source for each element. "List every item in the image and what it's based on."

Question surprises. If something looks off, ask where it came from, as I did.

Fill the gaps yourself. Give the AI your real stress buster (mine isn't a dog!) so it has no reason to guess.

Treat the output as a creative draft. It's entertainment and inspiration, not a fact sheet about you.

Final thoughts

The dog was harmless and kind of funny, but it showed me how AI works: it doesn't say "I don't know." It completes the picture. Our job is to notice what it invented and decide what to keep.

Have you caught an AI inventing something about you? Tell me in the comments.